def save(self):
"""
Saves the session.
This is done by saving a "session file" containing the cookies and any
other credientials, as well as a cache file containing the API data.
"""
session_file_path = config.CONFIG["session-path"]
if self.user:
logger.info("Saving session to %s.", session_file_path)
try:
if utils.prepare_directory(os.path.dirname(session_file_path)):
logger.info(
"Created directory(s) %s.",
os.path.dirname(session_file_path)
)
with open(session_file_path, "w") as f:
os.chmod(session_file_path, 0o600)
pickle.dump(
(
self.user,
requests.utils.dict_from_cookiejar(
self.requests_session.cookies
)
),
f
)
except IOError:
logger.warn(
"Could not save session to %s.",
session_file_path,
exc_info = sys.exc_info()
)
api_info_file_path = config.CONFIG["api-info-path"]
if self.api_info_raw:
logger.info("Saving API info to %s.", api_info_file_path)
try:
if utils.prepare_directory(os.path.dirname(api_info_file_path)):
logger.info(
"Created directory(s) %s.",
os.path.dirname(api_info_file_path)
)
with open(api_info_file_path, "wb") as f:
f.write(self.api_info_raw)
except IOError:
logger.warn(
"Could not save API Info to %s.",
api_info_file_path,
exc_info = sys.exc_info()
)
def main():
folders = glob.glob(os.path.join(args.glob_root, '*'))
folders = [f for f in folders if os.path.isdir(f) and 'combined-results' not in f]
print ('Experiment folders under {}:'.format(args.glob_root))
for folder in folders: print ('\t{}'.format(folder))
keys = args.keys.split(',') # names of the log attributes in log files
print ('Attributes whose values will be combined:', keys)
max_key_len = max([len(k) for k in keys])
combined_logs = {}
for k in keys: combined_logs[k] = []
utils.prepare_directory(args.save_dir, force_delete=True)
logger = utils.Logger(
args.save_dir,
'results',
list(map(lambda k: '{}/mean'.format(k), keys)) + \
list(map(lambda k: '{}/std'.format(k), keys)))
for folder in folders:
log_file = os.path.join(folder, args.log_file) # log file in the experiment folder
if not os.path.exists(log_file): continue
print ('\tLoading log file {} ...'.format(log_file))
L = np.load(log_file)
for k in keys:
v = L[k]
print ('\t\tkey:{key:{width}s}, value.shape:{keyshape:15s}, value[-1]:{keyval:}'.format(
key=k, width=max_key_len+1, keyshape=str(np.asarray(v).shape), keyval=v[-1]))
if len(v.shape) < 2:
v = v.reshape([-1, 1])
combined_logs[k].append(v)
for k in keys:
combined_k = np.concatenate(combined_logs[k], axis=1)
mean_k = combined_k.mean(axis=1)
std_k = combined_k.std(axis=1)
for lix, (mean, std) in enumerate(zip(mean_k, std_k)):
logger.append(['{}/mean'.format(k), '{}/std'.format(k)], [mean, std], lix)
print ('\taverage {key:{width}s}: {mean:} +- {std:}'.format(
key=k, width=max_key_len+1, mean=mean_k[-1], std=std_k[-1]))
logger.close()
def save(self):
"""
Saves the session.
This is done by saving a "session file" containing the cookies and any
other credientials, as well as a cache file containing the API data.
"""
session_file_path = config.CONFIG["session-path"]
if self.user:
logger.info("Saving session to %s.", session_file_path)
try:
if utils.prepare_directory(os.path.dirname(session_file_path)):
logger.info("Created directory(s) %s.",
os.path.dirname(session_file_path))
with open(session_file_path, "w") as f:
os.chmod(session_file_path, 0o600)
pickle.dump((self.user,
requests.utils.dict_from_cookiejar(
self.requests_session.cookies)), f)
except IOError:
logger.warn("Could not save session to %s.",
session_file_path,
exc_info=sys.exc_info())
api_info_file_path = config.CONFIG["api-info-path"]
if self.api_info_raw:
logger.info("Saving API info to %s.", api_info_file_path)
try:
if utils.prepare_directory(
os.path.dirname(api_info_file_path)):
logger.info("Created directory(s) %s.",
os.path.dirname(api_info_file_path))
with open(api_info_file_path, "wb") as f:
f.write(self.api_info_raw)
except IOError:
logger.warn("Could not save API Info to %s.",
api_info_file_path,
exc_info=sys.exc_info())
parser.add_argument('--data_dir', type=str, default='')
parser.add_argument('--exp_dir', type=str, default='./test/')
parser.add_argument('--ckpt_file', type=str, default='')
parser.add_argument('--device', type=str, default='cuda')
# parser.add_argument('--multi_gpu', action='store_true')
parser.add_argument('--dataset', type=str, default='Simp')
parser.add_argument('--test_split', type=float, default=0.2)
parser.add_argument('--red_rate', type=float, default=0.0)
parser.add_argument('--validation_split', type=float, default=0.0)
parser.add_argument('--d_latent', type=int, default=256)
parser.add_argument('--batch_size', type=int, default=32)
parser.add_argument('--n_epochs', type=int, default=2000)
parser.add_argument('--size', type=int, default=256)
parser.add_argument('--logfile', type=str, default='test.txt')
args = parser.parse_args()
if args.device == 'cuda' and torch.cuda.is_available():
from subprocess import call
print('available gpus:')
call([
"nvidia-smi", "--format=csv",
"--query-gpu=index,name,driver_version,memory.total,memory.used,memory.free"
])
cudnn.benchmark = True
else:
args.device = 'cpu'
utils.prepare_directory(args.exp_dir)
utils.write_logs(args)
configure(args.exp_dir)
main(args)
def _download(self, url, file_name):
"""
See :meth:`download`.
"""
downloads_directory = config.CONFIG["downloads-directory"]
if utils.prepare_directory(downloads_directory):
logger.info(
"Created directory(s) %s.",
downloads_directory
)
# Find an available file path
final_file_path = utils.find_available_file(
os.path.join(downloads_directory, file_name)
)
final_file_name = os.path.basename(final_file_path)
logger.debug("File will be saved to %s.", final_file_path)
# Get a generator function that makes a pretty progress bar.
bar = ui.progress_bar_indeterminate()
# Actually try to grab the file from the server
while True:
ui.print_carriage(
"%s Trying to download file... %s" %
(ui.progress_bar(0.0), " " * 30)
)
# Ask the server for the file
try:
file_request = self.requests_session.get(
url, timeout = 1, stream = True, verify = _get_verify()
)
except requests.exceptions.Timeout:
logger.info(
"Request timed out. Server did not accept connection after "
"1 second."
)
# If it's giving it to us...
if file_request.status_code == 200:
logger.debug(
"Response headers...\n%s",
pprint.pformat(file_request.headers, width = 72)
)
if "content-length" in file_request.headers:
size = float(file_request.headers["content-length"])
else:
logger.info("File is of unknown size.")
size = 0
ui.print_carriage(
ui.progress_bar(-1) + " Downloading file."
)
# Download the file in chunks.
chunk_size = 124
downloaded = 0
with open(final_file_path, "wb") as f:
for chunk in file_request.iter_content(124):
if size != 0:
ui.print_carriage(
ui.progress_bar(downloaded / size) +
" Downloading file."
)
downloaded += chunk_size
f.write(chunk)
# If the server got particularly angry at us...
if (file_request.status_code == 500 or
("X-CallSuccess" in file_request.headers and
file_request.headers["X-CallSuccess"] == "False")):
logger.critical(
"500 response. The server encountered an error."
)
sys.exit(1)
if file_request.status_code == requests.codes.ok:
break
# Make sure that the trying prompt appears for at least a moment or
# so
time.sleep(0.5)
period = 0.1
wait_for = 4
for i in xrange(int(wait_for / period)):
ui.print_carriage(
next(bar) + " Download not ready yet. Waiting."
)
time.sleep(period)
print "File saved to %s." % utils.shorten_path(final_file_path)
choices=['cuda', 'cpu'])
# classifier arguments
parser.add_argument('--lr', type=float, default=0.)
parser.add_argument('--wd', type=float, default=0.)
parser.add_argument('--batch_size', type=int, default=0)
parser.add_argument('--n_epoch', type=int, default=0)
args = parser.parse_args()
np.set_printoptions(linewidth=150, precision=4, suppress=True)
th.set_printoptions(linewidth=150, precision=4)
FN = th.from_numpy
join = os.path.join
logger = logging.getLogger()
utils.prepare_directory(args.exp_root, force_delete=True)
utils.init_logger(join(args.exp_root, 'program.log'))
utils.write_args(args)
dset = data.XianDataset(args.data_dir,
args.mode,
feature_norm=args.feature_norm)
_X_s_tr = FN(dset.X_s_tr).to(args.device)
_Y_s_tr = FN(dset.Y_s_tr).to(args.device)
_X_s_te = FN(dset.X_s_te).to(args.device)
_Y_s_te = FN(dset.Y_s_te).to(args.device)
_X_u_te = FN(dset.X_u_te).to(args.device)
_Y_u_te = FN(dset.Y_u_te).to(args.device)
_Cu = FN(dset.Cu).to(args.device)
_Sall = FN(dset.Sall).to(args.device)
def main():
utils.prepare_directory(args.exp_dir, force_delete=False)
utils.init_logger(join(args.exp_dir, 'program.log'))
utils.write_args(args)
# **************************************** load dataset ****************************************
dset = data.XianDataset(args.data_dir,
args.mode,
feature_norm=args.feature_norm)
_X_s_tr = FN(dset.X_s_tr).to(args.device)
_Y_s_tr_ix = FN(dil(dset.Y_s_tr,
dset.Cs)).to(args.device) # indexed labels
_Ss = FN(dset.Sall[dset.Cs]).to(args.device)
_Su = FN(dset.Sall[dset.Cu]).to(args.device)
if args.d_noise == 0: args.d_noise = dset.d_attr
# **************************************** create data loaders ****************************************
_sampling_weights = None
if args.dataset != 'SUN':
_sampling_weights = data.compute_sampling_weights(
dil(dset.Y_s_tr, dset.Cs)).to(args.device)
xy_iter = data.Iterator([_X_s_tr, _Y_s_tr_ix],
args.batch_size,
sampling_weights=_sampling_weights)
label_iter = data.Iterator([torch.arange(dset.n_Cs, device=args.device)],
args.batch_size)
class_iter = data.Iterator([torch.arange(dset.n_Cs)], 1)
# **************************************** per-class means and stds ****************************************
# per class samplers and first 2 class moments
per_class_iters = []
Xs_tr_mean, Xs_tr_std = [], []
Xs_te_mean, Xs_te_std = [], []
Xu_te_mean, Xu_te_std = [], []
for c_ix, c in enumerate(dset.Cs):
# training samples of seen classes
_inds = np.where(dset.Y_s_tr == c)[0]
assert _inds.shape[0] > 0
_X = dset.X_s_tr[_inds]
Xs_tr_mean.append(_X.mean(axis=0, keepdims=True))
Xs_tr_std.append(_X.std(axis=0, keepdims=True))
if args.n_gm_iter > 0:
_y = np.ones([_inds.shape[0]], np.int64) * c_ix
per_class_iters.append(
data.Iterator([FN(_X).to(args.device),
FN(_y).to(args.device)],
args.per_class_batch_size))
# test samples of seen classes
_inds = np.where(dset.Y_s_te == c)[0]
assert _inds.shape[0] > 0
_X = dset.X_s_te[_inds]
Xs_te_mean.append(_X.mean(axis=0, keepdims=True))
Xs_te_std.append(_X.std(axis=0, keepdims=True))
# test samples of unseen classes
for c_ix, c in enumerate(dset.Cu):
_inds = np.where(dset.Y_u_te == c)[0]
assert _inds.shape[0] > 0
_X = dset.X_u_te[_inds]
Xu_te_mean.append(_X.mean(axis=0, keepdims=True))
Xu_te_std.append(_X.std(axis=0, keepdims=True))
del _X, _inds, c_ix, c
Xs_tr_mean = FN(np.concatenate(Xs_tr_mean, axis=0)).to(args.device)
Xs_tr_std = FN(np.concatenate(Xs_tr_std, axis=0)).to(args.device)
Xs_te_mean = FN(np.concatenate(Xs_te_mean, axis=0)).to(args.device)
Xs_te_std = FN(np.concatenate(Xs_te_std, axis=0)).to(args.device)
Xu_te_mean = FN(np.concatenate(Xu_te_mean, axis=0)).to(args.device)
Xu_te_std = FN(np.concatenate(Xu_te_std, axis=0)).to(args.device)
# **************************************** create networks ****************************************
g_net = modules.get_generator(args.gen_type)(
dset.d_attr, args.d_noise, args.n_g_hlayer, args.n_g_hunit,
args.normalize_noise, args.dp_g, args.leakiness_g).to(args.device)
g_optim = optim.Adam(g_net.parameters(),
args.gan_optim_lr_g,
betas=(args.gan_optim_beta1, args.gan_optim_beta2),
weight_decay=args.gan_optim_wd)
d_net = modules.ConditionalDiscriminator(dset.d_attr, args.n_d_hlayer,
args.n_d_hunit,
args.d_normalize_ft, args.dp_d,
args.leakiness_d).to(args.device)
d_optim = optim.Adam(d_net.parameters(),
args.gan_optim_lr_d,
betas=(args.gan_optim_beta1, args.gan_optim_beta2),
weight_decay=args.gan_optim_wd)
start_it = 1
utils.model_info(g_net, 'g_net', args.exp_dir)
utils.model_info(d_net, 'd_net', args.exp_dir)
if args.n_gm_iter > 0:
if args.clf_type == 'bilinear-comp':
clf = classifiers.BilinearCompatibility(dset.d_ft, dset.d_attr,
args)
elif args.clf_type == 'mlp':
clf = classifiers.MLP(dset.d_ft, dset.n_Cs, args)
utils.model_info(clf.net, 'clf', args.exp_dir)
pret_clf = None
if os.path.isfile(args.pretrained_clf_ckpt):
logger.info('Loading pre-trained {} checkpoint at {} ...'.format(
args.clf_type, args.pretrained_clf_ckpt))
ckpt = torch.load(args.pretrained_clf_ckpt, map_location=args.device)
pret_clf = classifiers.BilinearCompatibility(dset.d_ft, dset.d_attr,
args)
pret_clf.net.load_state_dict(ckpt[args.clf_type])
pret_clf.net.eval()
for p in pret_clf.net.parameters():
p.requires_grad = False
pret_regg = None
if os.path.isfile(args.pretrained_regg_ckpt):
logger.info(
'Loading pre-trained regressor checkpoint at {} ...'.format(
args.pretrained_regg_ckpt))
ckpt = torch.load(args.pretrained_regg_ckpt, map_location=args.device)
pret_regg = classifiers.Regressor(args, dset.d_ft, dset.d_attr)
pret_regg.net.load_state_dict(ckpt['regressor'])
pret_regg.net.eval()
for p in pret_regg.net.parameters():
p.requires_grad = False
training_log_titles = [
'd/loss',
'd/real',
'd/fake',
'd/penalty',
'gm/loss',
'gm/real_loss',
'gm/fake_loss',
'g/fcls_loss',
'g/cycle_loss',
'clf/train_loss',
'clf/train_acc',
'mmad/X_s_tr',
'mmad/X_s_te',
'mmad/X_u_te',
'smad/X_s_tr',
'smad/X_s_te',
'smad/X_u_te',
]
if args.n_gm_iter > 0:
training_log_titles.extend([
'grad-cossim/{}'.format(n) for n, p in clf.net.named_parameters()
])
training_log_titles.extend(
['grad-mse/{}'.format(n) for n, p in clf.net.named_parameters()])
training_logger = utils.Logger(os.path.join(args.exp_dir, 'training-logs'),
'logs', training_log_titles)
t0 = time.time()
logger.info('penguenler olmesin')
for it in range(start_it, args.n_iter + 1):
# **************************************** Discriminator updates ****************************************
for p in d_net.parameters():
p.requires_grad = True
for p in g_net.parameters():
p.requires_grad = False
for _ in range(args.n_d_iter):
x_real, y_ix = next(xy_iter)
s = _Ss[y_ix]
x_fake = g_net(s)
d_real = d_net(x_real, s).mean()
d_fake = d_net(x_fake, s).mean()
d_penalty = modules.gradient_penalty(d_net, x_real, x_fake, s)
d_loss = d_fake - d_real + args.L * d_penalty
d_optim.zero_grad()
d_loss.backward()
d_optim.step()
training_logger.update_meters(
['d/real', 'd/fake', 'd/loss', 'd/penalty'], [
d_real.mean().item(),
d_fake.mean().item(),
d_loss.item(),
d_penalty.item()
], x_real.size(0))
# **************************************** Generator updates ****************************************
for p in d_net.parameters():
p.requires_grad = False
for p in g_net.parameters():
p.requires_grad = True
g_optim.zero_grad()
[y_fake] = next(label_iter)
s = _Ss[y_fake]
x_fake = g_net(s)
# wgan loss
d_fake = d_net(x_fake, s).mean()
g_wganloss = -d_fake
# f-cls loss
fcls_loss = 0.0
if pret_clf is not None:
fcls_loss = pret_clf.loss(x_fake, _Ss, y_fake)
training_logger.update_meters(['g/fcls_loss'], [fcls_loss.item()],
x_fake.size(0))
# cycle-loss
cycle_loss = 0.0
if pret_regg is not None:
cycle_loss = pret_regg.loss(x_fake, s)
training_logger.update_meters(['g/cycle_loss'],
[cycle_loss.item()], x_fake.size(0))
g_loss = args.C * fcls_loss + args.R * cycle_loss + g_wganloss
g_loss.backward()
# gmn iterations
for _ in range(args.n_gm_iter):
c = next(class_iter)[0].item()
x_real, y_real = next(per_class_iters[c])
y_fake = y_real.detach().repeat(args.gm_fake_repeat)
s = _Ss[y_fake]
x_fake = g_net(s)
# gm loss
clf.net.zero_grad()
if args.clf_type == 'bilinear-comp':
real_loss = clf.loss(x_real, _Ss, y_real)
fake_loss = clf.loss(x_fake, _Ss, y_fake)
elif args.clf_type == 'mlp':
real_loss = clf.loss(x_real, y_real)
fake_loss = clf.loss(x_fake, y_fake)
grad_cossim = []
grad_mse = []
for n, p in clf.net.named_parameters():
# if len(p.shape) == 1: continue
real_grad = grad([real_loss], [p],
create_graph=True,
only_inputs=True)[0]
fake_grad = grad([fake_loss], [p],
create_graph=True,
only_inputs=True)[0]
if len(p.shape) > 1:
_cossim = F.cosine_similarity(fake_grad, real_grad,
dim=1).mean()
else:
_cossim = F.cosine_similarity(fake_grad, real_grad, dim=0)
# _cossim = F.cosine_similarity(fake_grad, real_grad, dim=1).mean()
_mse = F.mse_loss(fake_grad, real_grad)
grad_cossim.append(_cossim)
grad_mse.append(_mse)
training_logger.update_meters(
['grad-cossim/{}'.format(n), 'grad-mse/{}'.format(n)],
[_cossim.item(), _mse.item()], x_real.size(0))
grad_cossim = torch.stack(grad_cossim)
grad_mse = torch.stack(grad_mse)
gm_loss = (1.0 -
grad_cossim).sum() * args.Q + grad_mse.sum() * args.Z
gm_loss.backward()
training_logger.update_meters(
['gm/real_loss', 'gm/fake_loss'],
[real_loss.item(), fake_loss.item()], x_real.size(0))
g_optim.step()
# **************************************** Classifier update ****************************************
if args.n_gm_iter > 0:
if it % args.clf_reset_iter == 0:
if args.clf_reset_iter == 1:
# no need to generate optimizer each time
clf.init_params()
else:
clf.reset()
else:
x, y_ix = next(xy_iter)
if args.clf_type == 'bilinear-comp':
clf_acc, clf_loss = clf.train_step(x, _Ss, y_ix)
else:
clf_acc, clf_loss = clf.train_step(x, y_ix)
training_logger.update_meters(
['clf/train_loss', 'clf/train_acc'], [clf_loss, clf_acc],
x.size(0))
# **************************************** Log ****************************************
if it % 1000 == 0:
g_net.eval()
# synthesize samples for seen classes and compute their first 2 moments
Xs_fake_mean, Xs_fake_std = [], []
with torch.no_grad():
for c in range(dset.n_Cs):
y = torch.ones(256, device=args.device,
dtype=torch.long) * c
a = _Ss[y]
x_fake = g_net(a)
Xs_fake_mean.append(x_fake.mean(dim=0, keepdim=True))
Xs_fake_std.append(x_fake.std(dim=0, keepdim=True))
Xs_fake_mean = torch.cat(Xs_fake_mean)
Xs_fake_std = torch.cat(Xs_fake_std)
# synthesize samples for unseen classes and compute their first 2 moments
def compute_firsttwo_moments(S, C):
X_mean, X_std = [], []
with torch.no_grad():
for c in range(dset.n_Cu):
y = torch.ones(
256, device=args.device, dtype=torch.long) * c
a = _Su[y]
x_fake = g_net(a)
X_mean.append(x_fake.mean(dim=0, keepdim=True))
X_std.append(x_fake.std(dim=0, keepdim=True))
X_mean = torch.cat(X_mean)
X_std = torch.cat(X_std)
Xu_fake_mean, Xu_fake_std = [], []
with torch.no_grad():
for c in range(dset.n_Cu):
y = torch.ones(256, device=args.device,
dtype=torch.long) * c
a = _Su[y]
x_fake = g_net(a)
Xu_fake_mean.append(x_fake.mean(dim=0, keepdim=True))
Xu_fake_std.append(x_fake.std(dim=0, keepdim=True))
Xu_fake_mean = torch.cat(Xu_fake_mean)
Xu_fake_std = torch.cat(Xu_fake_std)
g_net.train()
training_logger.update_meters([
'mmad/X_s_tr', 'smad/X_s_tr', 'mmad/X_s_te', 'smad/X_s_te',
'mmad/X_u_te', 'smad/X_u_te'
], [
torch.abs(Xs_tr_mean - Xs_fake_mean).sum(dim=1).mean().item(),
torch.abs(Xs_tr_std - Xs_fake_std).sum(dim=1).mean().item(),
torch.abs(Xs_te_mean - Xs_fake_mean).sum(dim=1).mean().item(),
torch.abs(Xs_te_std - Xs_fake_std).sum(dim=1).mean().item(),
torch.abs(Xu_te_mean - Xu_fake_mean).sum(dim=1).mean().item(),
torch.abs(Xu_te_std - Xu_fake_std).sum(dim=1).mean().item()
])
training_logger.flush_meters(it)
elapsed = time.time() - t0
per_iter = elapsed / it
apprx_rem = (args.n_iter - it) * per_iter
logging.info('Iter:{:06d}/{:06d}, '\
'[ET:{:.1e}(min)], ' \
'[IT:{:.1f}(ms)], ' \
'[REM:{:.1e}(min)]'.format(
it, args.n_iter, elapsed / 60., per_iter * 1000., apprx_rem / 60))
if it % 10000 == 0:
utils.save_checkpoint(
{
'g_net': g_net.state_dict(),
'd_net': d_net.state_dict(),
'g_optim': g_optim.state_dict(),
'd_optim': d_optim.state_dict(),
'iteration': it
},
args.exp_dir,
None,
it if it % (args.n_iter // args.n_ckpt) == 0 else None,
)
training_logger.close()
def _download(self, url, file_name):
"""
See :meth:`download`.
"""
downloads_directory = config.CONFIG["downloads-directory"]
if utils.prepare_directory(downloads_directory):
logger.info("Created directory(s) %s.", downloads_directory)
# Find an available file path
final_file_path = utils.find_available_file(
os.path.join(downloads_directory, file_name))
final_file_name = os.path.basename(final_file_path)
logger.debug("File will be saved to %s.", final_file_path)
# Get a generator function that makes a pretty progress bar.
bar = ui.progress_bar_indeterminate()
# Actually try to grab the file from the server
while True:
ui.print_carriage("%s Trying to download file... %s" %
(ui.progress_bar(0.0), " " * 30))
# Ask the server for the file
try:
file_request = self.requests_session.get(url,
timeout=1,
stream=True,
verify=_get_verify())
except requests.exceptions.Timeout:
logger.info(
"Request timed out. Server did not accept connection after "
"1 second.")
# If it's giving it to us...
if file_request.status_code == 200:
logger.debug("Response headers...\n%s",
pprint.pformat(file_request.headers, width=72))
if "content-length" in file_request.headers:
size = float(file_request.headers["content-length"])
else:
logger.info("File is of unknown size.")
size = 0
ui.print_carriage(
ui.progress_bar(-1) + " Downloading file.")
# Download the file in chunks.
chunk_size = 124
downloaded = 0
with open(final_file_path, "wb") as f:
for chunk in file_request.iter_content(124):
if size != 0:
ui.print_carriage(
ui.progress_bar(downloaded / size) +
" Downloading file.")
downloaded += chunk_size
f.write(chunk)
# If the server got particularly angry at us...
if (file_request.status_code == 500
or ("X-CallSuccess" in file_request.headers
and file_request.headers["X-CallSuccess"] == "False")):
logger.critical(
"500 response. The server encountered an error.")
sys.exit(1)
if file_request.status_code == requests.codes.ok:
break
# Make sure that the trying prompt appears for at least a moment or
# so
time.sleep(0.5)
period = 0.1
wait_for = 4
for i in xrange(int(wait_for / period)):
ui.print_carriage(
next(bar) + " Download not ready yet. Waiting.")
time.sleep(period)
print "File saved to %s." % utils.shorten_path(final_file_path)